Detection Method of Insulation Gloves Wearing in Complex Scenes Based on Improved YOLOX

Supervision of wearing insulation gloves during electrical equipment inspection is the top priority of safety protection in indoor and outdoor high-voltage power areas. However, in the actual scene, there are usually incomplete features of monitoring targets and insufficient feature information of small-scale targets caused by factors such as occlusion and distance, which leads to low accuracy of inspection personnel wearing insulation gloves. In view of the above situation, this paper proposes an improved testing model for insulation gloves. Firstly, SCAM attention mechanism and M-MHSA module were integrated into the feature extraction network to improve the ability of the model to extract global information and target channel features combined with multi-head attention mechanism. Adding small target detection layer and using weighted bidirectional feature pyramid network (BiFPN) to carry out multi-scale feature fusion can improve the detection ability of the model on different scale targets. The experimental results show that the improved algorithm achieves 93.27% average accuracy and 32frame/s detection speed in indoor and outdoor high-voltage electricity usage scenarios, which has good performance in the detection task of insulation gloves.